US8805763B2 - Computerized system for monitoring and controlling physical data-producing apparatus - Google Patents
Computerized system for monitoring and controlling physical data-producing apparatus Download PDFInfo
- Publication number
- US8805763B2 US8805763B2 US13/977,344 US201113977344A US8805763B2 US 8805763 B2 US8805763 B2 US 8805763B2 US 201113977344 A US201113977344 A US 201113977344A US 8805763 B2 US8805763 B2 US 8805763B2
- Authority
- US
- United States
- Prior art keywords
- data
- points
- physical
- physical data
- anomalous
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/02—Knowledge representation; Symbolic representation
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B23/00—Testing or monitoring of control systems or parts thereof
- G05B23/02—Electric testing or monitoring
- G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
- G05B23/0218—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults
- G05B23/0243—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults model based detection method, e.g. first-principles knowledge model
- G05B23/0254—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults model based detection method, e.g. first-principles knowledge model based on a quantitative model, e.g. mathematical relationships between inputs and outputs; functions: observer, Kalman filter, residual calculation, Neural Networks
-
- G06K9/053—
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
- G06F2218/10—Feature extraction by analysing the shape of a waveform, e.g. extracting parameters relating to peaks
Definitions
- the present invention relates generally to control systems and more particularly to feedback-providing control systems.
- PDM Phase Dispersion Minimalization
- Certain embodiments of the present invention seek to provide a method for finding, fitting and subtracting periodic signals in order to identify and supply a real time alert on anomalous episodes.
- Certain embodiments of the present invention seek to provide a method for finding one or more periodicities of data e.g. using the Power Spectrum (Fourier Transform) or the PDM (Phase Dispersion Minimalization) technique, fitting the data with this period e.g. using the least squares method, and subtracting the fit from the data to show anomalies e.g. by utilizing the standard deviation of the residual data to define or determine the significance of an anomalous data-point.
- the Power Spectrum Freourier Transform
- PDM Phase Dispersion Minimalization
- Certain embodiments of the present invention seek to provide a method for finding anomalies and supplying a suitable alert, e.g. in real time, each anomaly including a sequence of anomalous points such that the probability of randomly obtaining such an anomaly are cumulative and thus very low.
- Certain embodiments of the present invention are applicable to data with multiple periodicities, multiple semi-periods or a combination of both, and to data in which the periodic variation appears less than all the time, e.g. daily between 8-10 am.
- a method wherein the monitoring includes subtracting each predicted data point from a corresponding physical data point thereby to generate a sequence of residual data points.
- a method comprising generating the predetermined information from previously physical data points.
- a method comprising displaying the residual data points to a user.
- the periodic variations include multi-periodic variations which are taken into account when identifying wayward points.
- a method wherein at least one period characterizing the periodic variations is found using a phase dispersion minimalization method.
- the physical data comprises data generated by a satellite.
- the physical data comprises GPS system data.
- a method wherein the physical data comprises north-finding system data.
- a method wherein the physical data comprises particle accelerator data.
- a method wherein the physical data comprises cardiac data.
- a method wherein the physical data comprises brain activity (EEG) data.
- EEG brain activity
- a method wherein the physical data comprises atomic clock data.
- a method wherein the physical data comprises oscillator data.
- a method wherein the physical data comprises pendulum data.
- the physical data comprises solar energy data having periodicity depending on the sun's periodicity.
- a method wherein the physical data comprises temperature data.
- a method wherein the physical data comprises electrical equipment data.
- a method wherein the physical data comprises computer component data.
- a computerized system for monitoring physical data for anomalies, which physical data are predictable given predetermined information comprising: a predetermined information data point generator operative to compute a sequence of model data points which the physical data, given predetermined information, can be expected to duplicate at each of a corresponding sequence of temporal sampling points; a wayward point monitor including a processor operative for monitoring the physical data including identifying wayward points within the physical data that are incongruous with the predetermined information; and, typically, an anomalous episode-prompted alarm generator operative for identifying anomalous episodes each including a cluster of wayward points satisfying predefined anomalous episode-defining criteria and generating, for each anomalous episode identified, an alarm including, typically, the statistical significance of the anomalous episode.
- a method wherein the physical system action comprises computerized generation of an anomaly event alert for a human operator.
- the physical system action comprises a computerized change in at least one operating parameter of equipment associated with the physical data.
- a method wherein the physical system action comprises computerized suspension of operation of equipment associated with the physical data, due to an anomaly alert.
- a method wherein the physical system action comprises a computerized initiation of a data back-up process, for data associated with the physical data.
- a method wherein the physical system action comprises computerized launching of an emergency procedure.
- a method wherein the physical system action comprises computerized initiation of re-calibration for equipment associated with the data.
- a method wherein the physical system action comprises computerized initiation of at least one predetermined test to diagnose reasons for anomaly.
- a method wherein the characteristics comprises a probability of at least the individual anomalous episode occurring by chance.
- a method wherein the physical system action comprises computerized generation of an anomaly event alert for an external system.
- the physical data comprises ambient temperature data generated by the satellite.
- a method comprising using the physical data generator to generate the physical data generator.
- a method comprising identifying as wayward points, those points within the sequence of residual data points whose divergence from zero conforms to a waywardness criterion.
- a method comprising identifying as wayward points, those points within the sequence of residual data points which diverge, at a level exceeding a threshold value, from the standard deviation of the residual data.
- a method wherein the threshold value is determined based on the number of points in the physical data.
- the physical data includes N points and wherein the monitoring includes subtracting each predicted data point from a corresponding physical data point thereby to generate a sequence of residual data points defining noise and identifying as wayward points, those points which diverge from noise by M standard deviations of the residual data, where M is higher than expected given N.
- the residual data points each define a divergence from zero and wherein the waywardness criterion comprises divergence from zero, of a point within the sequence of residual data points, which is larger, by a threshold amount, than most other residual data points' divergences from zero.
- a computer program product comprising a computer usable medium having a computer readable program code embodied therein, the computer readable program code adapted to be executed to implement any of the methods shown and described herein.
- the data are visually inspected to decide which of the methods shown herein to employ, or whether to employ a suitable combination of such methods.
- the visual inspection may for example be employed by a human user to determine how many periods to use and whether there is a semi-period, or perhaps a few, and which method to employ.
- the visual inspection step may be wholly or partially automated.
- the fit is effected simultaneously:
- the raw data are fitted with n periods simultaneously so as to obtain the best fit.
- the method does not, therefore, say, fit p1, subtract it, and fit p2 to the residual data, but instead, the method typically includes finding p1, initially subtracting it from the residual data so as to conclude that there is also p2, and then simultaneously fitting p1 and p2 to the raw data for a better fit. Nonetheless, in certain applications, the first, non-simultaneous, alternative may be employed.
- n an integer
- D data, a 2-dimensional vector (2*n matrix) with n points, where the first column is time (in any units) and the second is any parameter (in any units).
- M computational model
- a model may include the earth's one-day rotational period as well as the earth's one-year period.
- M 1 , M 12 . . . first model (fitting one period), second model (fitting first and second periods) . . . .
- Multi-periodic data which have several periods
- Semi-period a periodical element or component in varying data which is not purely periodic.
- data are periodic, but period varies over time, e.g. as in some electrocardiograms
- data are periodic, but amplitude within each period varies systematically over time, e.g. as in some electrocardiograms
- data are periodic, but phase within each period varies systematically over time, e.g. as in some electrocardiograms
- Semi-frequency inverse of the period parameter of a semi-period (1/period)
- a computer program product comprising a computer usable medium or computer readable storage medium, typically tangible, having a computer readable program code embodied therein, the computer readable program code adapted to be executed to implement any or all of the methods shown and described herein. It is appreciated that any or all of the computational steps shown and described herein may be computer-implemented. The operations in accordance with the teachings herein may be performed by a computer specially constructed for the desired purposes or by a general purpose computer specially configured for the desired purpose by a computer program stored in a computer readable storage medium.
- Any suitable processor, display and input means may be used to process, display e.g. on a computer screen or other computer output device, store, and accept information such as information used by or generated by any of the methods and apparatus shown and described herein; the above processor, display and input means including computer programs, in accordance with some or all of the embodiments of the present invention.
- processors workstation or other programmable device or computer or electronic computing device, either general-purpose or specifically constructed, used for processing; a computer display screen and/or printer and/or speaker for displaying; machine-readable memory such as optical disks, CDROMs, magnetic-optical discs or other discs; RAMs, ROMs, EPROMs, EEPROMs, magnetic or optical or other cards, for storing, and keyboard or mouse for accepting.
- processor includes a single processing unit or a plurality of distributed or remote such units.
- the above devices may communicate via any conventional wired or wireless digital communication means, e.g. via a wired or cellular telephone network or a computer network such as the Internet.
- the apparatus of the present invention may include, according to certain embodiments of the invention, machine readable memory containing or otherwise storing a program of instructions which, when executed by the machine, implements some or all of the apparatus, methods, features and functionalities of the invention shown and described herein.
- the apparatus of the present invention may include, according to certain embodiments of the invention, a program as above which may be written in any conventional programming language, and optionally a machine for executing the program such as but not limited to a general purpose computer which may optionally be configured or activated in accordance with the teachings of the present invention. Any of the teachings incorporated herein may wherever suitable operate on signals representative of physical objects or substances.
- computer and the term “computerized” should be broadly construed to cover any kind of electronic device with data processing capabilities, including, by way of non-limiting example, personal computers, servers, computing systems, communication devices, processors (e.g. digital signal processor (DSP), microcontrollers, field programmable gate array (FPGA), application specific integrated circuit (ASIC), etc.) and other electronic computing devices or devices with an embedded processor.
- processors e.g. digital signal processor (DSP), microcontrollers, field programmable gate array (FPGA), application specific integrated circuit (ASIC), etc.
- DSP digital signal processor
- FPGA field programmable gate array
- ASIC application specific integrated circuit
- Any suitable input device such as but not limited to a sensor, may be used to generate or otherwise provide information received by the apparatus and methods shown and described herein.
- Any suitable output device or display may be used to display or output information generated by the apparatus and methods shown and described herein.
- Any suitable processor may be employed to compute or generate information as described herein e.g. by providing one or more modules in the processor to perform functionalities described herein.
- Any suitable computerized data storage e.g. computer memory may be used to store information received by or generated by the systems shown and described herein.
- Functionalities shown and described herein may be divided between a server computer and a plurality of client computers. These or any other computerized components shown and described herein may communicate between themselves via a suitable computer network.
- FIG. 1 is a simplified flowchart illustration of a method for computerized monitoring of a system to be monitored, using data with at least one periodic component, all in accordance with certain embodiments of the present invention.
- FIG. 2 is a simplified flowchart illustration of a method for computerized monitoring of a system to be monitored, using data having a single period generated by the system, all in accordance with certain embodiments of the present invention.
- FIG. 3A is a simplified flowchart illustration of a method for performing step 210 in the method of FIG. 2 , in accordance with certain embodiments of the present invention.
- FIG. 3B is a simplified flowchart illustration of a method for performing step 220 in the method of FIG. 2 , in accordance with certain embodiments of the present invention.
- FIG. 3C is a simplified flowchart illustration of a method for performing step 240 in the method of FIG. 2 , in accordance with certain embodiments of the present invention.
- FIG. 4A is a simplified flowchart illustration of a method for computerized monitoring of a system to be monitored, using multi-periodic data generated by the system, in accordance with certain embodiments of the present invention.
- FIG. 4B is a simplified flowchart illustration of a method for performing step 410 in the method of FIG. 4A , in accordance with certain embodiments of the present invention.
- FIG. 5 is a simplified flowchart illustration of a method for performing step 420 in the method of FIG. 4A , in accordance with certain embodiments of the present invention.
- FIG. 6 is a simplified flowchart illustration of a method for computerized monitoring of a system to be monitored, using semi-periodic data generated by the system, in accordance with certain embodiments of the present invention.
- FIG. 7A is a simplified flowchart illustration of a method for performing step 610 in the method of FIG. 6 , in accordance with certain embodiments of the present invention.
- FIG. 7B is a simplified flowchart illustration of a method for performing step 620 in the method of FIG. 6 , in accordance with certain embodiments of the present invention.
- FIG. 8A is a simplified flowchart illustration of a method for computerized monitoring of a system to be monitored, using multiple semi-periodic data (e.g. data, having multiple semi-periods) generated by the system, in accordance with certain embodiments of the present invention.
- multiple semi-periodic data e.g. data, having multiple semi-periods
- FIG. 8B is a simplified flowchart illustration of a method for performing step 810 in the method of FIG. 8A , in accordance with certain embodiments of the present invention.
- FIG. 9 is a simplified flowchart illustration of a method for performing step 820 in the method of FIG. 8A , in accordance with certain embodiments of the present invention.
- FIG. 10 is a simplified functional block diagram illustration of a computerized, data monitoring control system useful in performing certain methods shown and described herein such as the method of FIG. 1 .
- FIGS. 11A-19 are graphs illustrating an example of an application of the method of FIG. 2 to satellite data.
- FIGS. 20-24 are graphs illustrating examples of operation of steps 310 and 320 of FIG. 3A pertaining to visual inspection of data and period and semi-period finding, respectively.
- FIG. 25 is a set of 5 graphs respectively showing a data stream, first harmonic fit, residuals from the first harmonic fit, second harmonic fit and residuals from the second harmonic fit.
- FIG. 26 is a set of 4 graphs respectively showing a 10-harmonic fit to the data of FIG. 25 , the residuals thereof, a 20-harmonic fit to the data of FIG. 11A , and the residuals thereof.
- FIG. 27 is a set of 5 graphs respectively showing another data stream, first harmonic fit, residuals from the first harmonic fit, second harmonic fit and residuals from the second harmonic fit.
- FIG. 28 is a set of 4 graphs respectively showing a 10-harmonic fit to the data of FIG. 27 , the residuals thereof, a 20-harmonic fit to the data of FIG. 11A , and the residuals thereof.
- FIGS. 29-34 are graphs illustrating an example of an application of the method of FIG. 4A .
- FIG. 1 is a simplified flowchart illustration of a method for computerized monitoring of a system to be monitored, using data with at least one periodic component, all in accordance with certain embodiments of the present invention.
- the method of FIG. 1 typically comprises some or all of the following steps, suitably ordered e.g. as shown:
- Step 10 Finding the periodicity/ies of periodic, semi-periodic, multi-periodic, multiple semi-periodic or multi-periodic and multiple semi-periodic data e.g. conventionally
- Step 20 Fitting the data with the periodicity/ies and/or semi-periodicity/ies e.g. utilizing conventional data fitting techniques
- Step 30 Subtracting the fit from the data to generate residual data
- Step 40 In the residual data, identifying abnormal points that are above a certain level of standard deviations.
- Step 50 Declaring anomalous episodes, in which subsequent data-points largely deviate from many standard deviations of the residual data.
- Step 60 Estimating the probability of such events occurring by chance
- Step 70 Computerized production of an off-line or real time action (e.g. computerized generation of anomaly event alert for human operators or external systems, computerized change in operating parameters of equipment generating data, computer-initiated pause in equipment operation, computerized back-up, computerized launching of emergency procedure, computerized initiation of re-calibration for equipment generating data including automatic control equipment associated therewith, computerized initiation of predetermined tests to diagnose reasons for anomaly, computerized progress to a subsequent stage in a multi-stage procedure) as per a pre-determined schedule dependent on characteristics of anomalous episodes such as types found, quantity found, and probability of such events occurring by chance.
- an off-line or real time action e.g. computerized generation of anomaly event alert for human operators or external systems, computerized change in operating parameters of equipment generating data, computer-initiated pause in equipment operation, computerized back-up, computerized launching of emergency procedure, computerized initiation of re-calibration for equipment generating data including automatic control equipment associated therewith, computerized
- the physical operation/s performed in step 70 are application specific and depend on the physical apparatus generating the data; it is appreciated that the data may for example include satellite data, GPS system data, north-finding system data (an example of a North finding system is described in U.S. Pat. No. 5,060,392); particle accelerator data, cardiac data, brain data (EEG), atomic clock data; oscillator data; pendulum data; solar energy data having periodicity depending on the sun's periodicity; temperature data; electrical equipment data; computer component data.
- the alert may be accompanied by an indication of its statistical significance.
- decisions re operations to be initiated may be made in advance based inter alia on statistical significance. For example, if there is a problem with the temperature (say, 5 degrees too hot), with a significance level of 99.9999%, then shut down operation of satellite.
- Alert characteristics may indicate the significance e.g. an alarm that beeps louder when significance is higher.
- FIG. 2 is a simplified flowchart illustration of a method for identifying anomalies in data having a single period.
- FIG. 2 ( FIGS. 11A , 12 - 13 , 14 - 19 ).
- the method of FIG. 2 typically comprises some or all of the following steps, suitably ordered e.g. as shown: Step 210 .
- this preliminary step may be manually done once, although it can also be performed automatically. It is noted that usually the periodic nature of the data is known in advance, i.e. a 1-day period caused by the rotation of the Earth, or a 90-minutes satellite period etc. If the period is known and stable then this step can be omitted.
- Step 220 Build a model (M)
- Step 230 Subtract the model (also termed herein “fit”) from the data (D) and get the residual data (R).
- the time input is kept intact, and the residual data are the differences between the raw data points and the model at same time dots.
- Step 240 Identify anomalous points and episodes
- Step 250 computerized generation of physical response to anomalous points and episodes e.g. computerized generation of an anomaly alert, e.g. a late or a real time anomaly event alert.
- the statistical threshold for an alert can be set in advance according to application-specific requirements. For example, a client may request more alerts, say anything that is higher than 95% significance, in which case 1 out of 20 cases will beep on average. Or, the client may request less alerts, say anything above 99.9999% significance in which case 1 out of 1 million cases will beep on average.
- the model that fits existing data can clearly be extrapolated in time and yield predictions for future data and thus supply real time alerts.
- FIG. 3A is a simplified flowchart illustration of a method for performing step 210 in the method of FIG. 2 , in accordance with certain embodiments of the present invention.
- the method of FIG. 3 typically comprises some or all of the following steps, suitably ordered e.g. as shown:
- Step 310 A visual inspection of the data in order to get an impression of the presence of the period that modulates them.
- FIG. 11A or FIG. 14 top panel, may be employed to perform this step.
- visual inspection of the graph of FIG. 22 may yield that in addition to a 1-d period may be known to exist, there is a shorter period, of perhaps ⁇ 0.05 day that appears only between phases ⁇ 0.25- ⁇ 0.75 (rough estimate by eye) of the 1-day cycle.
- Step 320 Find the period/frequency using the Power Spectrum (Fourier Transform), the PDM (Phase Dispersion Minimalization) or any other technique. These methods can reveal periods that are missed by the naked eye. FIG. 15 may be used for this step.
- FIG. 3B is a simplified flowchart illustration of a method for performing step 220 in the method of FIG. 2 , in accordance with certain embodiments of the present invention.
- the method of FIG. 3B typically comprises some or all of the following steps, suitably ordered e.g. as shown:
- the number of harmonics can be determined as per application-specific criteria e.g. reduced for variations that have a sinusoidal-like shape and for faster computing times, or increased for a better fit and sufficient computing resources. So, the number of harmonics need not be 10, this value being stipulated herein merely by way of example. Generally, the number of harmonicsit may be suitably determined e.g. as per application-specific criteria as described above, at a preliminary stage or decided automatically according to the fit e.g. by stopping as soon as adding more harmonics only slightly improves the fit. Quality of fit may for example be quantified as STD of the raw data minus the model and may be compared to a fixed pre-defined threshold value or to a predetermined ratio of the STD of the raw data e.g. 1% as described herein.
- an alternative equivalent model can be constructed from the mean shape of the periodic variation, simply by folding the data onto the frequency. The model is then several identical cycles of this mean shape, which may change over time.
- the subtraction of the fit yields the STD (standard deviation) of the residual data.
- the process may be stopped once one goes below some limiting STD. For example:
- FIG. 14 (other than top panel) and 16 are relevant to this method of finding a suitable number of harmonics.
- a criterion for the presence of a linear trend can be determined in advance—i.e. use a trend only if the slope constant is larger that 0.001*A 1 .
- the decision can be manually made by a visual inspection of the data at step 1 . It is noted that the linear fit and the fit of the frequency and its harmonic can be separately done (i.e. subtracting first the linear trend and then fitting the residual data with the period), but a better fit would be achieved when the two are combined in a single fit.
- FIG. 3C is a simplified flowchart illustration of a method for performing step 240 in the method of FIG. 2 , in accordance with certain embodiments of the present invention.
- the method of FIG. 3C typically comprises some or all of the following steps, suitably ordered e.g. as shown:
- Step 360 Investigate the residual data, R, and identify abnormal points that are above its noise at a certain level of standard deviations (6) according to the normal (Gaussian) distribution.
- the threshold can be manually defined in advance or determined with a correlation to the number of data points to yield true alarms according to the normal distribution. In example, for 100 data points—the threshold can be set at about 2.2 ⁇ , for 1000 data points ⁇ 2.7 ⁇ or 3 ⁇ and for 1 million points— ⁇ 3.7 ⁇ or 4 ⁇ , etc.
- Step 370 Declare anomalous episodes, in which subsequent several data-points rise above the noise.
- the minimum number of points in an event can be manually defined in advance, in example as 1, 2, 10, n/100 etc. according to the data quality and client need.
- an “episode” may be defined by receiving an application-specific e.g. user-selected significance threshold and selecting a suitable number of points to “fit” that significance. The higher the requested significance, the less points/episodes/alerts are generated. In 100,000 points, 99.999% significance means a single point on average, whereas 99.5% means 500 points on average.
- FIG. 4A is a simplified flowchart illustration of a method for computerized monitoring of a system to be monitored, using multi-periodic data generated by the system, in accordance with certain embodiments of the present invention.
- the method of FIG. 4A typically comprises some or all of the following steps, suitably ordered e.g. as shown:
- FIGS. 29-34 multi-periodic ( FIGS. 29-34 ).
- the model employed may be:
- M 12 A 0 + C * ⁇ t + A 1 * ⁇ sin ⁇ ( 1 ⁇ w 1 ⁇ t + ⁇ 1 ⁇ - ⁇ 1 ) + A 2 * ⁇ sin ⁇ ( 2 ⁇ w 1 ⁇ t + ⁇ 1 ⁇ - ⁇ 2 ) + ... + A 10 * ⁇ sin ⁇ ( 10 ⁇ w 1 ⁇ t + ⁇ 1 ⁇ - ⁇ 10 ) ++ ⁇ B 1 * ⁇ sin ⁇ ( 1 ⁇ w 2 ⁇ t + ⁇ 2 ⁇ - ⁇ 1 ) + B 2 * ⁇ sin ⁇ ( 2 ⁇ w 2 ⁇ t + ⁇ 2 ⁇ - ⁇ 2 ) + ... + B 10 * ⁇ sin ⁇ ( 10 ⁇ w 2 ⁇ t + ⁇ 2 ⁇ - ⁇ 10 ) .
- Step 410 Find the periods (optional). For stable periods, i.e. periods which do not change over time, this preliminary step should be best manually done once, although it can also be performed automatically by fixing the maximum number of periods allowed for the model. It is noted that usually the periodic or multi-periodic nature of the data is known in advance, i.e. a 1-day period caused by the rotation of the Earth, a 90-minutes satellite period, a combination of both etc. If the periods are known and stable then method may proceed directly to step 590
- Step 420 Build the model (M)
- Step 430 Subtract the model from the data e.g. as described above with reference to FIG. 2 , step 230
- Step 440 Identify anomalous points and episodes e.g. as described above with reference to FIG. 3C
- Step 450 Produce an anomaly alert e.g. as described above with reference to FIG. 2 , step 250
- FIG. 4B is a simplified flowchart illustration of a method for performing step 410 in the method of FIG. 4A , in accordance with certain embodiments of the present invention.
- the method of FIG. 4B typically comprises some or all of the following steps, suitably ordered e.g. as shown:
- Step 480 A visual inspection of the data in order to get an impression of the presence of the periods that modulate them.
- Step 490 Find the first period/frequency using the Power Spectrum (Fourier Transform), the PDM (Phase Dispersion Minimalization) or any other technique. These methods can reveal periods that are missed by the naked eye.
- steps 480 and 490 may include, if desired, checking whether more than one e.g. several periods are present, and if so, skipping appropriate steps in FIG. 4A .
- 3 periods are identified by initial visual inspection of the data (human vision, computer vision). In this case, it is unnecessary and typically less advantageous, to find the first period (using the power spectrum), subtract it, re-run the power spectrum on the residuals, find the second period, find the fit of the two periods on the raw data, subtract this fit from the raw data, and re-run time the power spectrum on the new residual data (after the subtraction of the 2-periods fit), find the third period and find the 3-period fit to the raw data.
- the power spectrum may simply be run on the raw data and all 3 periods found at this stage, and then the 3-period fit to the raw data may be found.
- FIG. 5 is a simplified flowchart illustration of a method for performing step 420 in the method of FIG. 4A , in accordance with certain embodiments of the present invention.
- the method of FIG. 5 typically comprises some or all of the following steps, suitably ordered e.g. as shown:
- the number of harmonics can be reduced for variations that have a sinusoidal-like shape and for faster computing times, or increased for a better fit and sufficient computing resources.
- the optimal number of harmonics can be determined at a preliminary stage or decided automatically according to the fit—simply stop once adding more harmonics only slightly improves the fit.
- an alternative equivalent model can be constructed from the mean shape of the periodic variation, simply by folding the data onto the frequency. The model is then several identical cycles of this mean shape, which may change over time.
- a criterion for the presence of a linear trend can be determined in advance—i.e. use a trend only if the slope constant is larger that 0.001*A 1 .
- the decision can be manually made by a visual inspection of the data at step 1 . It is noted that the linear fit and the fit of the frequency and its harmonic can be separately done (i.e. subtracting first the linear trend and then fitting the residual data with the period), but a better fit would be achieved when the two are combined in a single fit.
- the time input is kept intact, and the residual data are the differences between the raw data points and the first model at same time dots.
- Step 560 (optional) Investigate the residual data, R 1 and decide whether the data are modulated with a second period and find its value. This can be done by a visual inspection and by performing the methods listed in step 2 above, but this time on the residual data, R 1 , rather than the raw data, D.
- Step 570 Fit the raw data, D, simultaneously with the two frequencies and their 10 harmonics, and optionally a linear trend:
- M 12 A 0 + C * ⁇ t + A 1 * ⁇ sin ⁇ ( 1 ⁇ w 1 ⁇ t + ⁇ 1 ⁇ - ⁇ 1 ) + A 2 * ⁇ sin ⁇ ( 2 ⁇ w 1 ⁇ t + ⁇ 1 ⁇ - ⁇ 2 ) + ... + A 10 * ⁇ sin ⁇ ( 10 ⁇ w 1 ⁇ t + ⁇ 1 ⁇ - ⁇ 10 ) ++ ⁇ B 1 * ⁇ sin ⁇ ( 1 ⁇ w 2 ⁇ t + ⁇ 2 ⁇ - ⁇ 1 ) + B 2 * ⁇ sin ⁇ ( 2 ⁇ w 2 ⁇ t + ⁇ 2 ⁇ - ⁇ 2 ) + ... + B 10 * ⁇ sin ⁇ ( 10 ⁇ w 2 ⁇ t + ⁇ 2 ⁇ - ⁇ 10 ) + Comment: the best fit is done simultaneously.
- a successful fit should be defined according to the standard deviation of the residual data, or to the ratio between the standard deviation of the residual data and the standard deviation of the raw data.
- Step 590 Fit the data simultaneously with all n periods, which were found in the data, and their 10 harmonics, and optionally a linear trend:
- FIG. 6 is a simplified flowchart illustration of a method for computerized monitoring of a system to be monitored, using semi-periodic data generated by the system, e.g. as described below with reference to FIGS. 20-24 , in accordance with certain embodiments of the present invention.
- the method of FIG. 6 typically comprises some or all of the following steps, suitably ordered e.g. as shown:
- Step 610 Find the semi-period (optional) e.g. as per FIG. 7A
- This preliminary step should be best manually done once, although it can also be performed automatically once the nature of the semi period is known—for example if the periodic variation appears only during certain time intervals of the day.
- This step and corresponding steps in other methods can be skipped, if the semi-period value and its semi-periodic nature (type of semi-period and parameters of that type) are known.
- This step can be performed automatically by pre-defining a type of semi-period to be identified in a particular application. For example, it may be desired to identify a semi-period that appears only during 3 successive hours of the day.
- the semi-period may be found automatically, particularly its type and parameters of its type are known in advance. For example, if the semi-period occurs only between 2-4 am, then a period search is conducted on only these data. However, a standard period search may also be fruitful even if the nature of the semi-period is unknown.
- Step 620 Build the model (M) e.g. as per FIG. 7B
- Step 630 Subtract the model from the data e.g. as per step 230 in FIG. 2
- Step 640 Identify anomalous points and episodes e.g. as per FIG. 3C
- Step 650 Produce an anomaly alert e.g. as per step 250 in FIG. 2
- FIG. 7A is a simplified flowchart illustration of a method for performing step 610 in the method of FIG. 6 , in accordance with certain embodiments of the present invention.
- the method of FIG. 7A typically comprises some or all of the following steps, suitably ordered e.g. as shown:
- Step 710 A visual inspection of the data to get an impression of the presence of the semi-period that modulates them and its nature.
- Step 720 Find the semi-period/semi-frequency using the Power Spectrum (Fourier Transform), the PDM (Phase Dispersion Minimalization) or any other known technique. This stage should be carefully done with an adjustment to the nature of the semi period. In example, if the periodic variation only appears between 2-4 pm, then only these data should be taken in the analysis—the search for the period. It is, however, noted that the value of the semi-period and its nature can also be automatically determined from the data using, in advance, some criteria for its nature.
- FIG. 7B is a simplified flowchart illustration of a method for performing step 620 in the method of FIG. 6 , in accordance with certain embodiments of the present invention.
- the method of FIG. 7B typically comprises some or all of the following steps, suitably ordered e.g. as shown:
- Step 730 Fit the data with the semi-frequency and its 10 harmonics. This step is similar to step 3 in FIG. 2 , but the fit should be done according to the nature of the semi-period. In example, if the periodic variation only appears between 2-4 pm, then the fit includes a periodic variation during these hours and a constant value for the rest of the day.
- Step 740 (optional) add a linear long-term trend (slope ⁇ C*t) to the fit if appropriate.
- a criterion for the presence of a linear trend can be also determined—i.e. use a trend only if the slope constant is larger that 0.001*A 1 .
- the decision can be manually done at step 1 . It is noted that the linear fit and the fit of the semi-frequency and its harmonic can be separately done, but a better fit would be achieved when the two are combined in a single fit.
- FIG. 8A is a simplified flowchart illustration of a method for computerized monitoring of a system to be monitored, using multiple semi-periodic data (e.g. data, having multiple semi-periods) generated by the system, in accordance with certain embodiments of the present invention.
- the method of FIG. 8A typically comprises some or all of the following steps, suitably ordered e.g. as shown:
- Step 810 Find the semi-periods (optional). If the semi-periods are known and stable then the method may proceed directly to FIG. 9 , step 930
- Step 820 Build the model (M)
- Step 830 Subtract the model from the data e.g. as described above with reference to FIG. 2 , step 230
- Step 840 Identify anomalous points and episodes e.g. as described above with reference to FIG. 3C
- Step 850 Produce an anomaly alert e.g. as described above with reference to FIG. 2 , step 250
- FIG. 8B is a simplified flowchart illustration of a method for performing step 810 in the method of FIG. 8A , in accordance with certain embodiments of the present invention.
- the method of FIG. 8B typically comprises some or all of the following steps, suitably ordered e.g. as shown:
- Step 880 A visual inspection of the data in order to get an impression of the presence of the semi-periods which modulate them.
- Step 890 Find the first semi-period/semi-frequency
- FIG. 9 is a simplified flowchart illustration of a method for performing step 820 in the method of FIG. 8A , in accordance with certain embodiments of the present invention.
- the method of FIG. 9 typically comprises some or all of the following steps, suitably ordered e.g. as shown:
- Step 930 Fit the data with the semi-frequency found in step 890 , and its 10 (e.g.) harmonics
- Step 940 (optional) add a linear long-term trend (slope ⁇ C*t) to the fit
- Step 950 Subtract the first fit (M I ) from the data and get the residual data after the subtraction of the first semi-frequency
- Step 960 (optional) Investigate the residual data, R 1 and decide whether the data are modulated with a second semi-period and find its value
- Step 970 Fit the raw data, D, simultaneously with the two semi-frequencies and their 10 harmonics, and optionally a linear trend
- Step 990 Fit the data simultaneously with all n semi-periods, which were found in the data, and their 10 (e.g.) harmonics, and optionally a linear trend
- one-period variations are typically handled, e.g. as per the method of FIG. 2 , before semi-periodic variations are handled, e.g. as per the method of FIG. 6 , unless the amplitudes of the former are very small compared with the amplitudes of the latter in which case semi-periodic variations may be handled first.
- the final fit is done for all periods and semi-periods together.
- FIG. 10 is a simplified functional block diagram illustration of a computerized data monitor system useful in performing certain methods shown and described herein such as the method of FIG. 1 .
- the system of FIG. 10 typically includes a Model data point generator 1010 which typically generates a sequence of predicted data points e.g., if a computational model is used for prediction, model data points.
- a physical data generator 1015 one possible example being a satellite—generates a corresponding sequence of physical data which typically includes a very long sequence of data which may include hundreds, millions or even hundreds of millions and more of data elements.
- a Wayward point monitor processor 1020 receives and processes the corresponding sequences and identifies wayward temporal locations (wayward points e.g.) at which the physical data deviates from the model data points.
- An episode seeker 1030 uses predefined e.g. system-defined or user-defined anomalous episode defining criteria to generate an output indication of anomalous episodes.
- a computerized anomalous episode-triggered generator 1040 of physical system actions generates at least one physical system action responsive to the output indication of anomalous episodes generated by episode seeker 1030 .
- the system action so generated may for example include one or more of the following: computerized generation of anomaly event alert for human operators; computerized generation of anomaly event alert for external systems, computerized change in operating parameters of equipment generating data, computer-initiated pause in equipment operation e.g. if episodes found are indicative of dangerous malfunction, computerized initiation of data back-up, computerized launching of emergency procedure e.g. if episodes found are indicative of danger which requires such procedures, computerized initiation of re-calibration for equipment generating data e.g. automatic control equipment associated therewith e.g. if episodes found are indicative of a need for re-calibration, computerized initiation of predetermined tests to diagnose reasons for anomaly, or computerized progress to a subsequent stage in a multi-stage procedure.
- the wayward points may be those at which the physical data strays to more than a predetermined extent, from the model data predicted by the model.
- the system of FIG. 10 may be operative to identify onset of one or more special events whose effect on the physical data is at least partly or even completely known to the system, but whose onset or onsets, are not known.
- the system might identify onset of occurrence of at least one route modification from a library of known route modifications which are partially or entirely known other than time of onset.
- the system may act upon detected onsets of special events, in accordance with a pre-determined logical schedule of physical actions contingent upon data analysis for identifying special events in accordance with the methods shown and described herein, mutatis mutandis.
- At least one predicted data point, of at least one of said special events may be subtracted from a corresponding physical data point thereby to generate a sequence of residual data points defining noise. Points which diverge from the noise by M standard deviations of the residual data, where M is higher than expected given N, may then be identified as wayward points.
- system of FIG. 10 may if desired be employed to operate upon a sequence of physical data which is not periodic but nonetheless is predictable in the sense that the data exhibits a known behavior which is not periodic.
- FIGS. 11A-19 are examples of an application of the method of FIG. 2 to satellite data.
- FIGS. 11A-19 are examples of an application of the method of FIG. 2 to satellite data.
- FIG. 11A is a graph of a first stream of data vs. time which is almost completely sinusoidal.
- FIG. 11B is a power spectrum (power vs. frequency) graph for the data of FIG. 11A showing a strong first harmonic at a one day period and a second weaker harmonic having a 0.5 day period
- FIG. 12 is a set of 5 graphs respectively showing the data of FIG. 11A , first harmonic fit, residuals from the first harmonic fit, second harmonic fit and residuals from the second harmonic fit.
- the standard deviation of the residual data is indicated, in this figure and in similar figures described herein, on the relevant panels (graphs).
- FIG. 13 is a set of 4 graphs respectively showing a 10-harmonic fit to the data of FIG. 11A , the residuals thereof, a 20-harmonic fit to the data of FIG. 11A , and the residuals thereof. It is appreciated that as harmonics are added, the standard deviation decreases, as shown.
- FIG. 14 is a set of 5 graphs respectively showing a second stream of data which is less sinusoidal than the data of FIG. 11A ; a first harmonic fit thereto which is less successful than the first harmonic fit of FIG. 12 , residuals from the first harmonic fit to the second stream of data, second harmonic fit to the second stream of data, and residuals from the second harmonic fit.
- FIG. 15 is a power spectrum (power vs. frequency) graph for the data of FIG. 14 showing a second harmonic which, since the data are less sinusoidal, is much stronger than the second harmonic of FIG. 11B , relative to the first harmonic.
- FIG. 16 is a set of 4 graphs respectively showing a 10-harmonic fit to the second data stream of FIG. 14 , the residuals thereof, a 20-harmonic fit to the data of FIG. 14 , and the residuals thereof. It is appreciated that as harmonics are added, the standard deviation decreases, as shown.
- FIG. 17 is a set of 5 graphs respectively showing a third stream of data; a first harmonic fit thereto, residuals from the first harmonic fit to the second stream of data, second harmonic fit to the fourth stream of data, and residuals from the second harmonic fit.
- FIG. 18 is a set of 4 graphs respectively showing a 10-harmonic fit to the third data stream of FIG. 17 , the residuals thereof, a 20-harmonic fit to the data of FIG. 17 , and the residuals thereof, which show a slope which remains after all 20 harmonics have been fitted. It is appreciated that as harmonics are added, the standard deviation decreases, as shown. The residual data still is characterized by a slope.
- FIG. 19 is a set of 5 graphs in which a first degree polynomial, with a slope as determined from the last graph of FIG. 18 , is first subtracted from the third data stream of FIG. 17 , and using the difference data, graphs of a 10-harmonic fit, the residuals thereof, a 20-harmonic fit, and the residuals thereof, are shown.
- the residual data have no slope and the final standard deviation (0.073, in the illustrated example) after subtracting the slope is significantly lower than without this term, as shown in FIG. 18 .
- FIGS. 20-24 are examples of an application of the method of FIG. 6 .
- FIGS. 20-24 are examples of an application of the method of FIG. 6 .
- FIG. 20 is a set of 5 graphs respectively showing a fourth data stream, first harmonic fit, residuals from the first harmonic fit, second harmonic fit and residuals from the second harmonic fit.
- FIG. 21 is a set of 4 graphs respectively showing a 10-harmonic fit to the fourth data stream of FIG. 20 , the residuals thereof, a 20-harmonic fit to the data of FIG. 17 , and the residuals thereof. It is appreciated that as harmonics are added, the standard deviation decreases as shown.
- FIG. 22 is a set of 4 graphs respectively showing a 10-harmonic fit to a fifth data stream, the residuals thereof, a 20-harmonic fit, and the residuals thereof.
- FIG. 23 is a set of 4 graphs respectively showing a 10-harmonic fit to a sixth data stream, the residuals thereof, a 20-harmonic fit, and the residuals thereof.
- FIG. 24 shows an enlarged view of the sixth data stream of FIG. 23 , showing that the data are modulated with a 1-day period, but in addition, there is also a semi-period that only appears during certain hours of the day and is inactive at other times.
- FIGS. 20-24 are examples of operation of steps 310 and 320 of FIG. 3A pertaining to visual inspection of data and period and semi-period finding, respectively.
- a semi period that only appears during certain cycles of a, say, 1-day period
- visual or automated inspection of FIGS. 23 and 24 may yield that the data has two semi periods, one at the beginning of the day, and one at the end.
- the mean shape of the 1-d variation may be subtracted to achieve a good fit.
- FIG. 25 is a set of 5 graphs respectively showing a seventh data stream, first harmonic fit, residuals from the first harmonic fit, second harmonic fit and residuals from the second harmonic fit. Two anomalous episodes in the first data sequence are clearly identifiable in the residuals from the second harmonic fit and are indicated by bubbles.
- FIG. 26 is a set of 4 graphs respectively showing a 10-harmonic fit to the data of FIG. 25 , the residuals thereof, a 20-harmonic fit to the data of FIG. 11A , and the residuals thereof.
- FIG. 27 is a set of 5 graphs respectively showing an eighth data stream, first harmonic fit, residuals from the first harmonic fit, second harmonic fit and residuals from the second harmonic fit. Two anomalous episodes in the first data sequence are clearly identifiable in the residuals from the second harmonic fit and are indicated by bubbles.
- FIG. 28 is a set of 4 graphs respectively showing a 10-harmonic fit to the data of FIG. 27 , the residuals thereof, a 20-harmonic fit to the data of FIG. 11A , and the residuals thereof.
- FIGS. 29-34 are an example of an application of the method of FIG. 4A .
- FIGS. 29-34 are an example of an application of the method of FIG. 4A .
- FIG. 29 is a pair of graphs including (top panel) a ninth data stream, which spans over a full year.
- a half-year period may be identified automatically or by a visual inspection, and the first harmonic fit is shown.
- the bottom panel shows a zoom into the first month, in which a one-day period is seen to occur.
- the presence of the half-year period is, of course, not apparent since the bottom graph's domain is only 30 days.
- FIG. 30 is a power spectrum (power vs. frequency) graph for the data of FIG. 29 , including a top graph with a one-day period and a bottom graph, which is a zoom into small frequencies, as shown on the left side of the top panel, with a half year period, which shows a large error.
- FIG. 31 is a graph of a 6-harmony fit for the two periods of the ninth data stream. It is appreciated that the long, half-year period is not entirely stable as is apparent from the truncated crest of the second peak (at about 220-300 days).
- FIG. 32 is an enlargement of the graph of FIG. 31 , over only 30 days vs. the one year domain of the graph of FIG. 31 .
- FIG. 33 is a set of graphs each showing 5 days-worth of the data of FIG. 31 . Adjacent graphs are 2 months apart.
- FIG. 34 is a graph of a sixth-harmonic residual of the ninth data stream; in which anomalous episodes are identifiable, although the residual has a greater variance than the residuals, say in FIGS. 12-14 and 16 .
- FIG. 29 corresponds to step 480 .
- the visual inspection of the data shows that there are two periods. The presence of the two periods is evident by a visual inspection. The presence of the second period may only be apparent a by visual inspection of the residuals or even only by analyzing the power spectrum of the residual data.
- FIG. 30 corresponds to step 490 of FIG. 4A . Inspection of the power spectrum shows that there are two periods, such that the residual data after the subtraction of the first period need not be checked in order to identify presence of the second period.
- FIG. 31 corresponds to step 570 , showing the fit.
- FIG. 34 corresponds to steps 430 and 440 in which the model is subtracted from the data and anomalous points and episodes are identified; significance computations are not shown.
- system monitoring based on at least semi-periodic data may be entirely computerized, e.g. by providing preliminary definitions of a maximum number of periods and/or semi-periods to use, and pre-defining range of allowed periods (e.g. 0.1-100 day) and/or maximum number of harmonics which may be used; and/or minimum threshold level of noise to achieve, and possibly other parameters.
- the periodic, semi-periodic, multi-periodic methods shown herein may be provided as user-selectable operation modes of a monitoring system in which case the user, based on visual inspection of the data, would select a suitable mode and would even parameterize e.g. might key in one or more of:
- the monitoring system may provide a “trial run” to check that a user-selected model works, before reverting to automatic monitoring mode.
- anomalous points when identifying wayward points, also termed herein “anomalous points”, the average of the residual data, which should be zero or very near zero for a poor fit, is typically not used as a criterion in order to find the anomalous points. Instead the standard deviation (STD) of the data is thresholded to identify these points. Since the residual points are typically normally distributed around zero, with positive as well as negative points, the more points available, the higher the chance to find points at the edge of the distribution, and this typically determines the threshold which is used to identify wayward points. Specifically, assuming normal distribution, about 70% of all points are found within one STD ( ⁇ 1 sigma-+1 sigma). About 99% of all points are found within ⁇ 2 STD ( ⁇ 2sigma-+2sigma), etc.
- STD standard deviation
- software components of the present invention including programs and data may, if desired, be implemented in ROM (read only memory) form including CD-ROMs, EPROMs and EEPROMs, or may be stored in any other suitable computer-readable medium such as but not limited to disks of various kinds, cards of various kinds and RAMs.
- ROM read only memory
- EEPROM electrically erasable programmable read-only memory
- Components described herein as software may, alternatively, be implemented wholly or partly in hardware, if desired, using conventional techniques.
- components described herein as hardware may, alternatively, be implemented wholly or partly in software, if desired, using conventional techniques.
- Any computer-readable or machine-readable media described herein is intended to include non-transitory computer- or machine-readable media.
- Any computations or other forms of analysis described herein may be performed by a suitable computerized method. Any step described herein may be computer-implemented.
- the invention shown and described herein may include (a) using a computerized method to identify a solution to any of the problems or for any of the objectives described herein, the solution optionally include at least one of a decision, an action, a product, a service or any other information described herein that impacts, in a positive manner, a problem or objectives described herein; and (b) outputting the solution.
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Automation & Control Theory (AREA)
- Data Mining & Analysis (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Alarm Systems (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
Description
M=A 0 +A 1*sin(1wt+Φ 1)+A 2*sin(2wt+Φ 2)+ . . . +A 10*sin(10wt+Φ 10).
M=A 0 +A 1*sin(1wt+Φ 1)+A 2*sin(2wt+Φ 2)+ . . . +A 10*sin(10wt+Φ 10)
M=A 0 +A 1*cos(1wt+Φ 1)+A 2*cos(2wt+Φ 2)+ . . . +A 10*cos(10wt+Φ 10)
or both the sine and cosine functions together, without the phases:
M=A 0 +A 1*sin(1wt)+B 1*cos(1wt)+ . . . +A 10*sin(10wt)+B 10*cos(10wt)
In addition, an alternative equivalent model can be constructed from the mean shape of the periodic variation, simply by folding the data onto the frequency. The model is then several identical cycles of this mean shape, which may change over time.
M=A 0 +C*t+A 1*sin(1wt+Φ 1)+A 2*sin(2wt+Φ 2)+ . . . +A 10*sin(10wt+Φ 10)
A criterion for the presence of a linear trend can be determined in advance—i.e. use a trend only if the slope constant is larger that 0.001*A1. Alternatively, the decision can be manually made by a visual inspection of the data at
M 1 =A 0 +A 1*sin(1w 1 t+Φ 1)+A 2*sin(2w 1 t+Φ 2)+ . . . +A 10*sin(10w 1 t+Φ 10)
The number of harmonics can be reduced for variations that have a sinusoidal-like shape and for faster computing times, or increased for a better fit and sufficient computing resources. The optimal number of harmonics can be determined at a preliminary stage or decided automatically according to the fit—simply stop once adding more harmonics only slightly improves the fit.
M 1 =A 0 +A 1*cos(1w 1 t+Φ 1)+A 2*cos(2w 1 t+Φ 2)+ . . . +A 10*cos(10w 1 t+Φ 10)
or both the sine and cosine functions together, without the phases:
M 1 =A 0 +A 1*sin(1w 1 t)+B 1*cos(1w 1 t)+ . . . +A 10*sin(10w 1 t)+A 10*cos(10w 1 t)
Also, an alternative equivalent model can be constructed from the mean shape of the periodic variation, simply by folding the data onto the frequency. The model is then several identical cycles of this mean shape, which may change over time.
M 1 =A 0 +C*t+A 1*sin(1w 1 t+Φ 1)+A 2*sin(2w 1 t+Φ 2)+ . . . +A 10*sin(10w 1 t+Φ 10)
A criterion for the presence of a linear trend can be determined in advance—i.e. use a trend only if the slope constant is larger that 0.001*A1. Alternatively, the decision can be manually made by a visual inspection of the data at
Comment: the best fit is done simultaneously.
Where the first three terms written out above are the first, second and tenth harmonics for a first period, the second three terms written out above are the first, second and tenth harmonics for a second period, and last three terms written out above are the first, second and tenth harmonics for an n'th period.
-
- If the data of
FIG. 11A is being used, a human user might key in information indicating that there is a period of ˜1 day, which is very sinusoidal, so few harmonics need be employed, and the monitoring system would select and parameterize a method accordingly. - If the data of
FIG. 14 are being used, a human user might key in information indicating that there is a period of ˜1 day, which is somewhat sinusoidal, so a medium number of harmonics may be employed. The monitoring system would select and parameterize a method accordingly. - If the data of
FIG. 18 (top panel) is being used, a human user might key in information indicating that there is a period of ˜1 day, which is somewhat sinusoidal, so a medium number of harmonics may be employed. A human user may also determine that a linear trend is present. The monitoring system may receive this information via a suitable user interface and may select and parameterize a method accordingly. - If the data of
FIGS. 20 , 22 and 23 (top panels) and ofFIG. 24 are being used, a human user might key in information indicating that there is a period of ˜1 day, and a semi-period of ˜0.1 day that appears during certain hours of the day. In example, a user's visual inspection ofFIG. 22 may show that in addition to the 1-d period, there is a shorter period (rough estimate by eye ˜0.05 day) that appears only between phases ˜0.25-˜0.75 (rough estimate by eye) of the 1-day cycle. The monitoring system would select and parameterize a method accordingly (e.g. would use the method ofFIG. 4A rather than, say, the method ofFIG. 2 ). - The following analysis may be performed: find the cycle intervals where the semi-period is active and check only these data. Compute the power spectrum to find the semi-period value, and then build the semi-period fit only for these intervals, using zero for the time intervals where it is inactive. This may be done automatically, if suitable parameters are defined in advance. Alternatively, check whether a mean shape of the 1-day period is a good fit to the data and subtract it to find anomalies.
- If the data of
FIG. 25 and ofFIG. 27 (top panel) are being used, a human user might key in information indicating that there is a period of ˜1 day, which is very sinusoidal, so few harmonics are needed. The monitoring system would select and parameterize a method accordingly. - If the data of
FIG. 29 is being used, a human user might key in information indicating that there are two periods of ˜1 day & ˜180 day. The monitoring system would select and parameterize a method accordingly.
- If the data of
-
- Example: Among a sample of 10 points, 3 points are expected within 1 sigma and 0.1 within 2 sigma, so even a 2-sigma point is typically considered highly significant hence may be considered wayward. However, among one million points a point that deviates by 3-sigma is not very rare, and a higher threshold, say 5 or 6 sigma, may be employed to identify wayward points.
Claims (27)
Applications Claiming Priority (3)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
IL210358A IL210358A (en) | 2010-12-29 | 2010-12-29 | Computerized system for monitoring and controlling physical data-producing apparatus |
IL210358 | 2010-12-29 | ||
PCT/IL2011/050067 WO2012090199A2 (en) | 2010-12-29 | 2011-12-19 | Computerized system for monitoring and controlling physical data-producing apparatus |
Publications (2)
Publication Number | Publication Date |
---|---|
US20130325785A1 US20130325785A1 (en) | 2013-12-05 |
US8805763B2 true US8805763B2 (en) | 2014-08-12 |
Family
ID=44262501
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
US13/977,344 Expired - Fee Related US8805763B2 (en) | 2010-12-29 | 2011-12-19 | Computerized system for monitoring and controlling physical data-producing apparatus |
Country Status (4)
Country | Link |
---|---|
US (1) | US8805763B2 (en) |
EP (1) | EP2659409B1 (en) |
IL (1) | IL210358A (en) |
WO (1) | WO2012090199A2 (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US11321164B2 (en) | 2020-06-29 | 2022-05-03 | International Business Machines Corporation | Anomaly recognition in information technology environments |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112380044B (en) * | 2020-12-04 | 2024-05-28 | 腾讯科技(深圳)有限公司 | Data anomaly detection method, device, computer equipment and storage medium |
Citations (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US4204225A (en) | 1978-05-16 | 1980-05-20 | Wisconsin Alumni Research Foundation | Real-time digital X-ray subtraction imaging |
US5060392A (en) | 1990-07-09 | 1991-10-29 | Allied-Signal Inc. | North finding system |
US20030191568A1 (en) | 2002-04-09 | 2003-10-09 | Breed David S. | Method and system for controlling a vehicle |
US20050186933A1 (en) | 1997-07-31 | 2005-08-25 | Francois Trans | Channel equalization system and method |
US20050192727A1 (en) | 1994-05-09 | 2005-09-01 | Automotive Technologies International Inc. | Sensor Assemblies |
US20060053136A1 (en) | 2004-08-09 | 2006-03-09 | Amir Ashiri | Method and system for analyzing multidimensional data |
US7222302B2 (en) * | 2003-06-05 | 2007-05-22 | International Business Machines Corporation | Method and apparatus for generating it level executable solution artifacts from the operational specification of a business |
US20070276270A1 (en) | 2006-05-24 | 2007-11-29 | Bao Tran | Mesh network stroke monitoring appliance |
US20080091309A1 (en) | 1998-01-15 | 2008-04-17 | Walker Richard C | Electrically controlled automated devices to operate, slow, guide, stop and secure, equipment and machinery for the purpose of controlling their unsafe, unattended, unauthorized, unlawful hazardous and/or legal use, with remote control and accountability worldwide |
US20090078875A1 (en) | 2004-11-09 | 2009-03-26 | Spectrum Dynamics Llc | Radioimaging |
US20090312998A1 (en) | 2006-07-06 | 2009-12-17 | Biorics Nv | Real-time monitoring and control of physical and arousal status of individual organisms |
US7724724B2 (en) * | 2003-11-07 | 2010-05-25 | Hitachi, Ltd. | Radio communication apparatus |
US7813822B1 (en) * | 2000-10-05 | 2010-10-12 | Hoffberg Steven M | Intelligent electronic appliance system and method |
US7885976B2 (en) * | 2007-02-23 | 2011-02-08 | International Business Machines Corporation | Identification, notification, and control of data access quantity and patterns |
US8080819B2 (en) * | 2004-07-08 | 2011-12-20 | Philips Solid-State Lighting Solutions, Inc. | LED package methods and systems |
US8207821B2 (en) * | 2003-05-05 | 2012-06-26 | Philips Solid-State Lighting Solutions, Inc. | Lighting methods and systems |
US8320831B2 (en) * | 2009-04-15 | 2012-11-27 | Kabushiki Kaisha Toshiba | Electronic device and communication control method |
Family Cites Families (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7024335B1 (en) * | 1998-04-15 | 2006-04-04 | The Texas A&M University System | Condition assessment and life expectancy prediction for devices |
FR2925182B1 (en) * | 2007-12-18 | 2021-07-02 | Airbus France | METHOD AND DEVICE FOR DETECTION OF OSCILLATORY FAILURES IN A SLAVE CHAIN IN POSITION OF AN AIRCRAFT RUDDER. |
GB0807775D0 (en) * | 2008-04-29 | 2008-06-04 | Romax Technology Ltd | Methods for model-based diagnosis of gearbox |
FR2939170B1 (en) * | 2008-11-28 | 2010-12-31 | Snecma | DETECTION OF ANOMALY IN AN AIRCRAFT ENGINE. |
-
2010
- 2010-12-29 IL IL210358A patent/IL210358A/en active IP Right Grant
-
2011
- 2011-12-19 EP EP11852387.7A patent/EP2659409B1/en active Active
- 2011-12-19 WO PCT/IL2011/050067 patent/WO2012090199A2/en active Application Filing
- 2011-12-19 US US13/977,344 patent/US8805763B2/en not_active Expired - Fee Related
Patent Citations (20)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US4204225A (en) | 1978-05-16 | 1980-05-20 | Wisconsin Alumni Research Foundation | Real-time digital X-ray subtraction imaging |
US5060392A (en) | 1990-07-09 | 1991-10-29 | Allied-Signal Inc. | North finding system |
US20050192727A1 (en) | 1994-05-09 | 2005-09-01 | Automotive Technologies International Inc. | Sensor Assemblies |
US20050186933A1 (en) | 1997-07-31 | 2005-08-25 | Francois Trans | Channel equalization system and method |
US20080091309A1 (en) | 1998-01-15 | 2008-04-17 | Walker Richard C | Electrically controlled automated devices to operate, slow, guide, stop and secure, equipment and machinery for the purpose of controlling their unsafe, unattended, unauthorized, unlawful hazardous and/or legal use, with remote control and accountability worldwide |
US7974714B2 (en) * | 1999-10-05 | 2011-07-05 | Steven Mark Hoffberg | Intelligent electronic appliance system and method |
US7813822B1 (en) * | 2000-10-05 | 2010-10-12 | Hoffberg Steven M | Intelligent electronic appliance system and method |
US20030191568A1 (en) | 2002-04-09 | 2003-10-09 | Breed David S. | Method and system for controlling a vehicle |
US8207821B2 (en) * | 2003-05-05 | 2012-06-26 | Philips Solid-State Lighting Solutions, Inc. | Lighting methods and systems |
US7222302B2 (en) * | 2003-06-05 | 2007-05-22 | International Business Machines Corporation | Method and apparatus for generating it level executable solution artifacts from the operational specification of a business |
US8094600B2 (en) * | 2003-11-07 | 2012-01-10 | Hitachi, Ltd. | Radio communication apparatus |
US7724724B2 (en) * | 2003-11-07 | 2010-05-25 | Hitachi, Ltd. | Radio communication apparatus |
US8116263B2 (en) * | 2003-11-07 | 2012-02-14 | Hitachi, Ltd. | Radio communication apparatus |
US8080819B2 (en) * | 2004-07-08 | 2011-12-20 | Philips Solid-State Lighting Solutions, Inc. | LED package methods and systems |
US20060053136A1 (en) | 2004-08-09 | 2006-03-09 | Amir Ashiri | Method and system for analyzing multidimensional data |
US20090078875A1 (en) | 2004-11-09 | 2009-03-26 | Spectrum Dynamics Llc | Radioimaging |
US20070276270A1 (en) | 2006-05-24 | 2007-11-29 | Bao Tran | Mesh network stroke monitoring appliance |
US20090312998A1 (en) | 2006-07-06 | 2009-12-17 | Biorics Nv | Real-time monitoring and control of physical and arousal status of individual organisms |
US7885976B2 (en) * | 2007-02-23 | 2011-02-08 | International Business Machines Corporation | Identification, notification, and control of data access quantity and patterns |
US8320831B2 (en) * | 2009-04-15 | 2012-11-27 | Kabushiki Kaisha Toshiba | Electronic device and communication control method |
Non-Patent Citations (2)
Title |
---|
Scargle, J.D., "Studies in Astronomical Time Series Analysis. II. Statistical Aspects of Spectral Analysis of Unevely Spaced Data," The Astrophysical Journal, Dec. 15, 1982, pp. 835-853, vol. 263. |
Stellingwerf, R.F., "Period Determination Using Phase Dispersion Minimization," The Astrophysical Journal, Sep. 15, 1978, pp. 953-960, vol. 224. |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US11321164B2 (en) | 2020-06-29 | 2022-05-03 | International Business Machines Corporation | Anomaly recognition in information technology environments |
Also Published As
Publication number | Publication date |
---|---|
IL210358A (en) | 2016-09-29 |
EP2659409A2 (en) | 2013-11-06 |
EP2659409A4 (en) | 2017-08-23 |
US20130325785A1 (en) | 2013-12-05 |
WO2012090199A3 (en) | 2013-10-17 |
IL210358A0 (en) | 2011-06-30 |
WO2012090199A2 (en) | 2012-07-05 |
EP2659409B1 (en) | 2019-03-20 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Davis et al. | LIGO detector characterization in the second and third observing runs | |
CN106104496B (en) | The abnormality detection not being subjected to supervision for arbitrary sequence | |
US20210334656A1 (en) | Computer-implemented method, computer program product and system for anomaly detection and/or predictive maintenance | |
CN110647414A (en) | Improved analog function security with anomaly detection | |
US11165799B2 (en) | Anomaly detection and processing for seasonal data | |
Couto et al. | Predicting software defects with causality tests | |
US8954362B2 (en) | System and method for predictive analytics in an electrical grid network | |
US20180005199A1 (en) | System and method to analyze and detect anomalies in vehicle service procedures | |
Hongxia et al. | Study on predictive maintenance strategy | |
Prakash et al. | A probabilistic approach to remaining useful life prediction of rolling element bearings | |
Pagano | A predictive maintenance model using long short-term memory neural networks and Bayesian inference | |
US8805763B2 (en) | Computerized system for monitoring and controlling physical data-producing apparatus | |
Ke et al. | Remaining useful life prediction for non-stationary degradation processes with shocks | |
Tamssaouet et al. | System-level failure prognostics: Literature review and main challenges | |
Xu et al. | Structural damage detection using low-rank matrix approximation and cointegration analysis | |
CN105825130A (en) | Information security early-warning method and device | |
Vagnoli et al. | An automatic bridge damage diagnostics method using empirical mode decomposition based health indicators and neuro‐fuzzy classification | |
Zonouz et al. | A fuzzy-Monte Carlo simulation approach for fault tree analysis | |
US9535917B1 (en) | Detection of anomalous utility usage | |
Fernández-Cantí et al. | Nonlinear set-membership identification and fault detection using a Bayesian framework: Application to the wind turbine benchmark | |
Yamada et al. | Software reliability | |
US20200133253A1 (en) | Industrial asset temporal anomaly detection with fault variable ranking | |
Nielsen | Value of information of structural health monitoring with temporally dependent observations | |
CN117121530A (en) | Composite event estimation through time logic | |
Muradova | Methods for managing the reliability and quality of IoT sensors |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
AS | Assignment |
Owner name: ISRAEL AEROSPACE INDUSTRIES LTD., ISRAEL Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:RETTER, ALON;REEL/FRAME:030745/0661 Effective date: 20120401 |
|
FEPP | Fee payment procedure |
Free format text: PAYOR NUMBER ASSIGNED (ORIGINAL EVENT CODE: ASPN); ENTITY STATUS OF PATENT OWNER: LARGE ENTITY |
|
STCF | Information on status: patent grant |
Free format text: PATENTED CASE |
|
FEPP | Fee payment procedure |
Free format text: SURCHARGE FOR LATE PAYMENT, LARGE ENTITY (ORIGINAL EVENT CODE: M1554) |
|
MAFP | Maintenance fee payment |
Free format text: PAYMENT OF MAINTENANCE FEE, 4TH YEAR, LARGE ENTITY (ORIGINAL EVENT CODE: M1551) Year of fee payment: 4 |
|
FEPP | Fee payment procedure |
Free format text: MAINTENANCE FEE REMINDER MAILED (ORIGINAL EVENT CODE: REM.); ENTITY STATUS OF PATENT OWNER: LARGE ENTITY |
|
LAPS | Lapse for failure to pay maintenance fees |
Free format text: PATENT EXPIRED FOR FAILURE TO PAY MAINTENANCE FEES (ORIGINAL EVENT CODE: EXP.); ENTITY STATUS OF PATENT OWNER: LARGE ENTITY |
|
STCH | Information on status: patent discontinuation |
Free format text: PATENT EXPIRED DUE TO NONPAYMENT OF MAINTENANCE FEES UNDER 37 CFR 1.362 |
|
FP | Lapsed due to failure to pay maintenance fee |
Effective date: 20220812 |